[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[netCDF #CRF-113718]: Garbage values in parallel netcdf write

I am not sure, but you are assuming that each call to nc_put_vara
will start up and continue writing after the nc_put_vara
returns. Thus doing 4 (#cores) calls to nc_put_vara
will cause each core to write its chunk in parallel with the others.
I do not know if this is what really happens.


In any case, I think you need to search the net to see
if you can find an example that does what you are trying to do.

Also, does this program work if you do not use mpi?


> No I’m not being clear enough, my bad.
> 
> The variable defined has dimensions
> [10][100][6][5][5].
> 
> If I run with four cores, I will make a single call to nc_put_vara_float, 
> with values
> start = (0, core_rank * 25, 0, 0, 0)
> count = (100, 25, 6, 5, 5)
> 
> where core_rank ranges from 0 to 3.
> 
> If it can help, attached is the c++ code that illustrates the problem. Notice 
> that I am using the Eigen library, because I wanted to reproduce the 
> situation I am encountering in my main project. I could also send you the 
> needed header files and makefile, if you'd like to run the code at some point.
> 
> Best,
> 
> Alex.
> 
> 
> 
> 
> 
> ________________________________
> From: Unidata netCDF Support <address@hidden>
> Sent: 13 September 2018 03:51:10
> To: Alexandre Szenicer
> Cc: address@hidden
> Subject: [netCDF #CRF-113718]: Garbage values in parallel netcdf write
> 
> Sorry, I am still unsure of what you are doing
> Are you doing a single call to nc_put_vara_float
> with the following values:
> start = (0,0,0,0,0)
> count = (10,100,6,5,5)
> and a chunk of memory of size
> sizeof(float) * 10 * 100 * 6 * 5 *5 bytes?
> 
> 
> 
> > Yes that is correct, I am using parallel netcdf, create the file with 
> > nc_open_par, and set the variable to collective access.
> > ________________________________
> > From: Unidata netCDF Support <address@hidden>
> > Sent: 12 September 2018 19:45:44
> > To: Alexandre Szenicer
> > Cc: address@hidden
> > Subject: [netCDF #CRF-113718]: Garbage values in parallel netcdf write
> >
> > Let me make sure I understand what you are doing.
> > When you say
> > > Each core has to write a vector of size [10][25][6][5][5]
> > do you mean that you are accessing the netcdf library in
> > a multi-threaded way?
> >
> > >
> > >
> > > I'm writing you with regards to an unexpected behaviour I encountered 
> > > when writing a variable in parallel with netcdf.
> > >
> > >
> > > The situation is the following. I have a 5 dimensional vector, say of 
> > > dimensions [10][100][6][5][5]. I want to write it in parallel with four 
> > > cores.
> > >
> > > Each core has to write a vector of size [10][25][6][5][5]. It is 
> > > necessary in my problem that the dimension split between the processes be 
> > > in second position.
> > >
> > > When this variable is written, it appears that there are a lot of 
> > > corrupted values in the resulting netcdf file.
> > >
> > > The alternative, which I am using at the moment, is to loop over the 
> > > first dimension, and perform write operations on 4D vectors instead. This 
> > > works fine, but becomes very time consuming for larger vectors, as I need 
> > > to perform many write operations.
> > >
> > >
> > > I am not sure if this behaviour is normal (something to do with 
> > > contiguity in memory) or not?
> > >
> > >
> > > Many thanks for your help.
> > >
> > >
> > > Best regards,
> > >
> > >
> > > Alexandre Szenicer.
> > >
> > >
> > >
> > >
> > >
> >
> > =Dennis Heimbigner
> > Unidata
> >
> >
> > Ticket Details
> > ===================
> > Ticket ID: CRF-113718
> > Department: Support netCDF
> > Priority: Normal
> > Status: Open
> > ===================
> > NOTE: All email exchanges with Unidata User Support are recorded in the 
> > Unidata inquiry tracking system and then made publicly available through 
> > the web.  If you do not want to have your interactions made available in 
> > this way, you must let us know in each email you send to us.
> >
> >
> >
> >
> 
> =Dennis Heimbigner
> Unidata
> 
> 
> Ticket Details
> ===================
> Ticket ID: CRF-113718
> Department: Support netCDF
> Priority: Normal
> Status: Open
> ===================
> NOTE: All email exchanges with Unidata User Support are recorded in the 
> Unidata inquiry tracking system and then made publicly available through the 
> web.  If you do not want to have your interactions made available in this 
> way, you must let us know in each email you send to us.
> 
> 
> 
> 

=Dennis Heimbigner
  Unidata


Ticket Details
===================
Ticket ID: CRF-113718
Department: Support netCDF
Priority: Normal
Status: Open
===================
NOTE: All email exchanges with Unidata User Support are recorded in the Unidata 
inquiry tracking system and then made publicly available through the web.  If 
you do not want to have your interactions made available in this way, you must 
let us know in each email you send to us.